1. Technical Field of the Invention
This invention relates to digital parallel processing systems, wherein a plurality of nodes communicate via messages over an interconnection network. In particular, this invention deals with the message store and forward mechanism which operates between each processing node and the interconnection network.
2. Background Art
In parallel systems, as also in PC clustering, a plurality of nodes communicate via messages sent over an interconnection network. Each node usually contains one processor and one network adapter to connect the processor to the network. State of the art network adapters provide memory on the adapter to store and forward messages to and from the network. The adapter memory usually becomes the bottleneck of the system. Consequently, there is a need in the art for a dynamic allocation method and apparatus capable of more efficient usage of the adapter memory.
Most network adapters in the art use a store and forward adapter memory to send and receive messages to and from the network. Normally a single memory at the adapter supports four ports to handle the message traffic to and from the network. Message traffic includes messages arriving at the adapter memory from the processor (Memory Port A), messages being sent from the adapter memory to the processor (Memory Port B), messages being sent from the adapter memory to the network (Memory Port C), and messages arriving, to the adapter memory from the network (Memory Port D). The prior art approach is to assign a static bandwidth to each of the 4 adapter memory ports to provide a balanced system. For instance, if the adapter memory supported a total bandwidth of 100 Mbytes/s, 25 Mbytes/s would be statically assigned to each port. This allocation is permanent and unchangeable. This type of approach assures that the memory will function successfully under worst case conditions, where all 4 ports are active simultaneously. However, the prior art approach reduces the performance of the system. A large percentage of the time the 4 ports are NOT active simultaneously, and the bandwidth associated with idle ports is wasted. Consequently, there is a need in the art for a method and system for dynamically allocating bandwidth to the adapter memory ports responsive to load conditions.
It is an object of the invention to provide a dynamic allocation method and apparatus capable of more efficient usage of the adapter memory.
It is a further object of the invention to provide a method and system for dynamically allocating bandwidth to the adapter memory ports responsive to load conditions.